視覺在生活當中佔了很重要的一部分，在這項研究中的視覺和識別可以改善輪型機器人（Kingdom National Instruments Robotics）之追蹤目標能力。本論文之研究方法是由兩種法則模糊理論與影像辨識整合而成。利用模糊理論建立模糊規則庫，設計出一個模糊滑動控制器，再將此演算法則應用在輪型機器人之目標追蹤上。此研究應用LabVIEW在影像處理上的優勢及控制上即時與精準性，完成本項成果。此項研究架構共分為兩個步驟，分述如下：
步驟一：以LabVIEW啟動輪型機器人(KNR)上的攝影機在固定時間內進行影像捕獲，並利用LabVIEW的視覺辨識功能對捕獲後的影像進行前置處理(二值化)與後置處理(影像辨識)。
步驟二：將模糊滑動控制器植入LabVIEW圖控程式中，計算輪型機器人左右輪轉速，降低輪型機器人(KNR)追蹤誤差。
本研究針對輪型機器人之目標追蹤，提出模糊理論及影像辨識方法，且已順利利用LabVIEW完成影像捕獲、前置處理、辨識及模糊滑動控制器之實現。本研究之基本成果，日後可再增強影像前置處理機制，以增強其辨識能力，也可大幅提升其實務應用，以解決影像辨識上之盲點，亦可提升影像追蹤目標物之精確性。
Vision plays an important role in daily life. The vision and recognition researches in this study can improve the target tracking performance of wheeled robot (Kingdom National Instruments Robotics). The used technology of this thesis is the combination of fuzzy theory and image processing. In fuzzy theory, the sliding surface, fuzzy rules, fuzzy inference and defuzzification are utilized to design a fuzzy sliding-mode controller (FSMC). The algorithm of the proposed FSMC is utilized to steer the wheeled robot and to track the target. To verify and implement the proposed architecture, all the program is developed under LabVIEW environment to calculate the speeds of right and left wheels, respectively. The developing stage contains following three steps:
Step 1: Develop the image capturing program under LabVIEW environment: The functions include (a) Triggering the camera of the controlled wheeled robots (KNR); (b) Capturing the image on fixed interval time write the Pre-processing (binary) and post-processing (pattern recognition) program to obtain the axis values.
Step 2: Design and implement the fuzzy sliding-mode controller: After defining the sliding surface, fuzzy set and fuzzy rules, the fuzzy inference algorithm is programmed under LabVIEW environment to calculate the speeds of right and left wheels, respectively.
To deal with the target tracking problem of the wheeled robot, the fuzzy theory and image processing methods are proposed in this study. And also all the proposed algorithms have been successfully executed under LabVIEW environment to capture the image, access the axis values and convert driving voltages to the motors of the wheeled robot (KNR). In the future, the performance can be further enhanced by embedding additional image pre-processing mechanisms to promote the recognition capability and to improve tracking accuracy of the robot in practical application.
